AI assistant, forecasting and anomaly detection

Talk to your data

The AI assistant answers questions such as "Which three lines had the most delays this week?" using your organization's data. The assistant sees only the data the user is authorized for.

• Daily summary: Highlights of yesterday's operations
• Forecast: Passenger demand and occupancy forecast per line
• Anomalies: Unusual increases in delays, occupancy or cost
• Suggestions: Capacity and assignment improvement suggestions, routed for approval

How does Copilot work?

To answer your question, Copilot uses RotaMerkezi's own tools, such as the line list, line occupancy, delay report, line comparison, anomalies, demand forecast and running scenarios. Each tool runs with your permissions and data scope; data you are not authorized for never reaches the assistant either. The answer lists which tools were used and links to the relevant scenario, suggestion or report.

• Answers appear on screen as they are written; long answers can be stopped.
• Numbers in the answer are checked against the tool results; any number that cannot be verified is corrected.
• Copilot makes no changes to live data. Scenarios and suggestions are only prepared; applying them requires the normal approval flow.
• No employee searches are performed; the assistant does not see employee addresses or contact details, and answers are generated from aggregate operational data.

Morning summary and tomorrow readiness

An automatic summary is prepared every morning and every evening: trip counts, first departure, expected passengers; trips without a vehicle or driver, over-capacity, drivers on leave, document status, open incidents, pending suggestions and yesterday's anomalies. The structure of the summary is deterministic; if you wish, the text is narrated by a language model and its numbers are verified.

Anomaly detection

Every night, the previous day's trips are examined. Ride time, delay and occupancy are compared with the typical values for the same line, direction and day type over the last 28 days. Unusual deviations are listed with their likely cause (route deviation, breakdown / accident, traffic, long waits at stops). Problems that recur over recent working days are flagged as "recurring".

Demand forecasting

Expected passengers per line are forecast for the next 7 days, based on assigned passengers, known leave and remote-work exceptions and (if boarding tracking is in place) historical no-show rates. If there is not enough historical data, no forecast is made up; "not enough data" is shown instead.

Providers and data protection

• The organization chooses between Anthropic (Claude) and OpenAI (GPT) and enters its own API key. The key is stored encrypted.
• The assistant cannot be enabled until consent for cross-border data transfer has been given.
• There is a monthly request quota and a per-user rate limit; usage and estimated cost are visible in the settings.

Frequently asked questions

What data does the AI assistant access?

Only operational data within the asking user's permissions and data scope, through RotaMerkezi's defined tools. No employee searches are performed; employee addresses and contact details are not given to the assistant.

Can the AI change the shuttle plan by itself?

No. The assistant can prepare scenarios and suggestions but does not write to live data. A change is applied only after being converted into a change set and approved by an authorized person.

Which AI models are supported?

Anthropic Claude and OpenAI GPT models. The organization chooses the provider and model and uses them with its own API key.

Do the summary and anomaly detection work without AI?

Yes. The morning summary, anomaly detection and demand forecasting are calculated statistically. The language model is used only for narrating the summary and for the Copilot chat.